Create a setup for running client/server jobs

[orting]

We should have a standardized way of running for example jupyter notebooks that use the cluster as backend. This would also be nice for things like interactive segmentation tools that are being developed at DIKU.

First step is to have a documented way of doing this.Then it would be nice with some easy to use scripts.

[orting]

Abe is using a simple approach with communication over files. The server is just started with slurm and the client is run locally with communication over sshfs mounting the relevant directory from the cluster locally. See https://github.com/Abe404/root_painter

[Ulfgard]

shouldn't we just be able to start a notebook on the node and use an ssh-tunnel to the root?

so just normal port-forwarding.

the problem is of course to ensure that the tunnel is working and that ports are free etc.

[Ulfgard]

here is a wordy guide describing how someone else did this:

https://alexanderlabwhoi.github.io/post/2019-03-08_jpn-slurm/

[orting]

Jupyter is one thing, and is pretty straightforward, except that you cant have fixed ports. We should have a documented way of doing this.
The more general case, client/server architecture with server running on the cluster can be solved in different ways. The benefit of using the file approach is that yo dont need to fiddle with ports, you just need access to a shared directory. I think it is much easier for people to do this.